← Dashboard

Google I/O 2026 — what actually matters

Straight talk: your I/O watcher's live pings silently failed through the keynote today (a character-encoding bug, fixed ~2pm). You missed the real-time alerts — that should not have happened. This page is the durable catch-up, written from what your curated channels actually published. Nothing here is fabricated; where I only had a headline, I say so.

The lede: two things from I/O actually move your stack this week. One, Google quietly slotted in a new $100/mo Gemini Ultra tier that undercuts ChatGPT Pro and Claude Max by half — same usage-ceiling pitch, with 20 TB Drive and YouTube Premium thrown in. Two, Gemini 3.5 Flash is now the default brain behind AI Mode in Search, which means a billion people just got pushed into agentic search whether they asked for it or not. Everything else is wallpaper around those two moves.

Gemini gets a Pro tier (and a plan for your money)

The pricing news is the one thing from this keynote that probably changes a decision you might actually make. Android Police walks through the new $100/mo Gemini Ultra tier: 5x the usage of the $20 Pro plan, 20 TB of Drive (vs 5 TB), YouTube Premium bundled, $40/mo of Google Cloud credit, 10,000 Flow Credits for the creative tools, and early access to Gemini Spark when it ships later this year. The old $250 super-tier dropped to $200 to clear room. The frame is unsubtle: aimed straight at $200 ChatGPT Pro and $200 Claude Max, at half price, with YouTube as the sweetener. Android Central reads the same plan as a productivity bundle — storage + cloud + creative credits, not just "more tokens."

We're all-in on Claude. This doesn't change that. But if you ever wanted a sandbox Google account with the actual Gemini ceiling unlocked (Bidet/TP3 eval, Veo experiments, Spark beta), $100 is the first time the math doesn't laugh.

The darker half of the pricing story: Android Police notes Google's CBO already floated porting Search's AI-Mode ad format into the Gemini app on the Q1 call. No announcement yet, but the door is officially open. A chatbot prompt is the richest targeting signal an advertiser has ever had, and the moment ads land in a "personal assistant" surface, every recommendation has to be read defensively. File and watch.

Search rebuilt around AI

This is the announcement Google framed as the biggest Search change in 25 years, and for once the hype is in the ballpark. Android Central lays out the actual mechanics: the search box now expands to a long-form input, accepts text plus images plus files plus video plus open Chrome tabs as context, and routes the whole thing through Gemini 3.5 Flash by default in AI Mode. AI Mode now has one billion monthly users with usage doubling each quarter, and Google is openly calling the keyword-search era over.

The model under all of this is Gemini 3.5 Flash, positioned as fast/cheap enough to make agents actually viable for high-volume tasks — with the bigger Pro model still weeks out. Read that as "Flash is the production workhorse, Pro is the demo darling." For TP3 the takeaway is mild: local Gemma 3 4B and Ollama still handle our extract/embed loop, our paid Gemini fallback is intentionally disabled, and nothing in this announcement changes that calculus. For everyone else who actually uses google.com? The search box they've used for 25 years is about to start talking back.

Generative media goes everywhere

Three media announcements, same idea: make generation a verb inside the app you were already using, not a separate "AI tool" you switch to.

Flow Music — Android Central flags a Flow Music app for generating or altering tunes and editing them on-device. Specs were thin (truncated piece) but the direction is Suno/Udio as a native Google app, on the phone, with editing baked in. Gemini Omni is the conversational video editor: describe a scene change, it does it. Light on availability.

Project Genie + Street View is the one that's actually weird in a good way. Android Central describes it as DeepMind's Genie world-model bolted onto real Street View: pick a US location, prompt it ("turn Chicago into a desert"), and you walk through an AI-altered version of the real place. Rolling out gradually to AI Ultra subscribers starting today, US-only. The one thing from the keynote that feels genuinely new instead of a feature checkbox.

Agents grow up — or try to

Google's pitch for the year is Gemini Spark, framed as a "24/7 personal AI agent" rather than a chatbot. Android Central covers the wedding-planning demo: Spark sits in the background across your Gmail/Docs/Slides, summarizes meetings, researches, coordinates threads — without re-prompting each step. Locked behind the $100 Ultra tier, ships "later in 2026." Third year running we've heard a version of this pitch, so wait for hands-on reviews before believing it.

The everyday-busywork wrapper around Spark is more interesting than the headline: Daily Brief assembles a personalized morning digest from Gmail + Calendar (sound familiar?), Gemini Spark for macOS targets local-file/desktop workflows, and Android Halo shows live agent activity as a system overlay so you can watch automations run without losing your screen. Halo is the one to watch — a real attempt at the "what is my agent doing right now" problem.

Docs Live is the most under-hyped useful thing. Android Police explains it: voice-drive Docs and Gmail conversationally — dump what you want, Gemini drafts, you keep talking to revise; same for Gmail replies and inbox search. Ships summer 2026, Pro/Ultra only. It's the exact Bidet brain-dump loop inside Google's surfaces — useful as a competitive reference point even if we never use it.

Anthropic / Claude lane (parallel)

I/O sucks the oxygen, but the Claude lane kept moving this week and a few items deserve top-billing.

Anthropic shipped a video on Project Glasswing, billed as an initiative to secure the world's software. Title-only (YouTube descriptions didn't fetch today) but worth the click on Anthropic's track record alone. Sam Witteveen has "How Claude's Design Agents Work" — his deep-dives on Claude tooling are consistently the best technical read in your radar.

On the indie/Claude-Code side, IndyDevDan posted a four-pack: "DELETE the BASH Tool: Agentic Security for Pi Agent and Claude Code" (start here — most directly relevant to how you run Claude Code on Apex), "GPT-5.5 vs Opus 4.7: A Pi Coding Agent That REVIEWS Like YOU", "My Pi Agent Teams. Claude Code Leak SIGNAL", and "One Agent Is NOT ENOUGH: Agentic Coding BEYOND Claude Code". Matthew Berman has three Claude takes: "So Anthropic is just winning now", "anthropic vs. openai", and "$400,000 from Claude Code" — the last is the kind of build-in-public proof point you keep flagging.

And a strange one: NLW covered a "Surprise Elon-Anthropic Team Up". No-Musk-products rule still holds, but if the headline is literal that's a strategic shift worth knowing the shape of. Note: Claire Vo / How I AI didn't surface anything new in this pull — her channel is in the source list now, so future pulls should catch her live.

What everyone's saying

NLW's pre-keynote framer is "What Google Needs to Do at I/O This Week"; worth pairing with his post-keynote take when it drops. AI Explained's "Gemini 3.1 Pro and the Downfall of Benchmarks: Welcome to the Vibe Era of AI" is the right argument even from the title — benchmarks have lost their meaning, "does this model feel smarter to me" is the actual eval.

Wes Roth has the developer-stream coverage (skip the "let's go" reaction). Two Minute Papers covered the DeepMind side — the research lens on the Genie/world-model story. TheoreticallyMedia has a Veo 4 leak + Krea vs Midjourney piece for the generative-video lane. 1littlecoder posted on OpenAI's GPT-Realtime-2 — real-time voice keeps inching forward; matters for the Ray-Bans north-star pattern. NLW also has "The Best Way to Talk to Your Agents" as a primer worth a click.

Skip these. The "did Google use a MacBook" piece is gossip. The Apple WWDC announcement is a date with no content. JulianGoldieSEO's "Gemini Updates Are Next Level" is SEO reaction; the Android Central pieces above cover it straight. The live blog is a wall of timestamps now — the synthesis above is the durable version. The "5 Gemini updates" listicle doesn't add anything the per-feature pieces don't already cover.

Sources actually fetched and synthesized: 8 Android Police / Android Central articles (Gemini Ultra pricing, Docs Live, ads risk, Search overhaul, Gemini 3.5 Flash, Project Genie/Street View, Gemini Spark, Flow Music, Gemini Omni, busywork features). YouTube descriptions were not retrievable today — YouTube items are listed by title with channel context only, called out where relevant.